Reading path: Coastal field methods →
Start with the observation, not the animal count
A passive acoustic monitor listens without sending out a sound of its own. A hydrophone converts pressure changes in the water into an electrical signal that can be recorded; marine animals, weather-driven processes and human activity can all contribute to an underwater soundscape. The first defensible statement is therefore about a recording: a specified signal was or was not detected by a specified system during a specified interval. [1] [2]
That is not automatically a count of animals. One animal may call repeatedly, several animals may call at once, and some animals may make no sound detectable by the chosen instrument. A call detection can support presence of that sound in the recorder’s acoustic range, but abundance, identity of an individual, or population change requires additional methods and assumptions.
Define what qualifies as a detection
Before processing audio, name the target: a particular call type, a frequency band, a sound-producing species, or a broader sound-level metric. These questions require different analyses. NOAA’s passive-acoustic archive guidance notes that the target signal’s frequency range and duration influence the detection method; a study may seek presence during an hour or day, or may preserve finer event timing. [3]
Write the rule in operational terms. For example, a hypothetical survey could classify an hour as positive when a trained reviewer confirms at least one call matching a documented call definition. That binary outcome answers whether the rule found a call in that hour. It does not report the number of callers or how often they called. If the question is call rate, the unit might instead be reviewed calls per usable recording hour, with a stated treatment of overlapping or uncertain events. The unit should match the question, and ambiguous clips should remain ambiguous rather than being forced into a species label.
For research explanations and other observation methods, browse science notes.
Make listening effort visible
A calendar day is not necessarily a full day of listening. Record the deployment and recovery times, actual recording schedule, file gaps, instrument uptime and any periods rejected during quality review. If one recorder supplies 24 usable hours and another supplies 15, raw daily detection totals are not comparable without accounting for that difference. A simple rate per usable hour may help describe effort, but it does not correct for all differences in detectability.
Acoustic availability also varies. Animals call at different times, signals have different frequency and duration, and sound travels through a changing environment before reaching a sensor. A non-detection can mean no target signal reached the sensor during the usable interval; it cannot by itself establish that no animal was present. [1]
Check the instrument and the sound path
For quantitative sound-level comparisons, the recording chain needs calibration information. NOAA NCEI recommends at least the hydrophone manufacturer specifications and prefers laboratory calibration before or after deployment. The archive also calls for details such as the analysis period, frequency range, processing method, protocol and software version. If calibration is unknown or sensitivity has degraded, NCEI cautions against using the resulting sound-level metrics for quantitative analysis. [3]
Detection workflows need their own checks: timestamps, sample rate, frequency response, microphone depth and orientation when known, noise conditions, and the software or listening protocol used. Likewise, a spectrogram can make patterns easier to inspect but is not a species identification by itself. Automated candidates should be checked against the stated call definition and, where feasible, reviewed by a person familiar with the target signals.
Compare records only after aligning their meaning
Suppose two coastal recorders each report detections in six of ten sampled hours. That equality says little if one sampled the same daytime window each day while the other sampled at night, or if their usable-hour totals, calibration, target definitions or review procedures differ. First harmonize the time zone, interval length, target rule, quality exclusions and effort denominator. Then show missing intervals rather than silently treating them as zero detections.
A sound-level summary and a call-detection table are not interchangeable. A sound-level metric can summarize energy across frequency bands, while an event detector asks whether a pattern meeting a rule occurred. NCEI notes that multiple signals may occupy overlapping frequency bands and that unbinned original audio may be needed to identify or separate signals. Keep the raw record or a traceable archive reference when possible, and label derived products with the processing steps that created them. [3]
Keep a record another analyst can audit
A compact acoustic field record should identify the project and site; coordinates and sensor depth; recorder and hydrophone model; serial number and calibration information; deployment and recovery times in UTC; sampling schedule and gaps; target signal and detection rule; analysis window; software and version; quality flags; and links to source audio and derived products. NOAA NCEI’s archive metadata guidance emphasizes the project, platform, hydrophone, sampling and analysis details needed to describe a passive-acoustic dataset. [3]
With those details, a detection becomes interpretable evidence rather than a mysterious mark on a map. It can support a careful question about when a sound was recorded, how consistently a site was sampled, or whether a soundscape changed under a defined method. Claims about animal numbers or ecological change should state what additional surveys, calibration, repeated sampling or detection models support them. A hydrophone is a powerful observer, but it observes through sound, effort and instrument limits.
Sources & further reading
Have a correction or a source to suggest? Send a note →